{"ID":2883010,"CreatedAt":"2026-06-01T04:54:23.091178241Z","UpdatedAt":"2026-06-01T04:54:23.091178241Z","DeletedAt":null,"paper_url":"https://arxiv.org/abs/2508.08547","arxiv_id":"2508.08547","title":"Calibration Attention: Learning Reliability-Aware Representations for Vision Transformers","abstract":"Most calibration methods operate at the logit level, implicitly assuming that miscalibration can be corrected without changing the underlying representation. We challenge this assumption and propose \\textbf{Calibration Attention (CalAttn)}, a \\emph{representation-aware} calibration module for vision transformers that couples instance-wise temperature scaling to transformer token geometry under a proper scoring objective. CalAttn predicts a sample-specific temperature from the \\texttt{[CLS]} token and backpropagates calibration gradients into the backbone, thereby reshaping the uncertainty structure of the representation rather than post-hoc adjusting confidence. This yields \\emph{token-conditioned uncertainty modulation} with negligible overhead (\\(\u003c0.1\\%\\) additional parameters). Across multiple datasets with ViT/DeiT/Swin backbones, CalAttn consistently improves calibration while preserving accuracy, achieving relative ECE reductions of \\(3.7\\%\\) to \\(77.7\\%\\) over strong baselines across diverse training objectives. Our results indicate that treating calibration as a representation-level problem is a practical and effective direction for trustworthy uncertainty estimation in transformers. Code: [https://github.com/EagleAdelaide/CalibrationAttention-CalAttn-](https://github.com/EagleAdelaide/CalibrationAttention-CalAttn-)","short_abstract":"Most calibration methods operate at the logit level, implicitly assuming that miscalibration can be corrected without changing the underlying representation. We challenge this assumption and propose \\textbf{Calibration Attention (CalAttn)}, a \\emph{representation-aware} calibration module for vision transformers that c...","url_abs":"https://arxiv.org/abs/2508.08547","url_pdf":"https://arxiv.org/pdf/2508.08547v2","authors":"[\"Wenhao Liang\",\"Wei Emma Zhang\",\"Lin Yue\",\"Miao Xu\",\"Mingyu Guo\",\"Olaf Maennel\",\"Weitong Chen\"]","published":"2025-08-12T01:19:12Z","proceeding":"cs.CV","tasks":"[\"cs.CV\"]","methods":"[\"Vision Transformer\",\"Transformer\"]","has_code":false,"code_links":[{"ID":610943,"CreatedAt":"2026-06-01T04:54:23.091178241Z","UpdatedAt":"2026-06-01T04:54:23.091178241Z","DeletedAt":null,"paper_id":2883010,"paper_url":"https://arxiv.org/abs/2508.08547","paper_title":"Calibration Attention: Learning Reliability-Aware Representations for Vision Transformers","repo_url":"https://github.com/EagleAdelaide/CalibrationAttention-CalAttn-","is_official":false,"mentioned_in_paper":false,"mentioned_in_github":true,"github_stars":0}]}